Automated versus manual detection of left ventricular wall motion abnormalities by gated blood pool SPECT

Abstract

The purpose of this investigation was to describe the rationale and implementation of new totally automated algorithms to compute global and regional left ventricular (LV) function measurements from blood pool (BP) gated SPECT data and to test the hypothesis that automated calculations are as accurate as manual calculations for detecting global and regional LV wall motion abnormalities when compared to independent cardiac magnetic resonance (CMR) measurements. From an Institutional Review Board approved retrospective review of data for 43 patients (age 58 +/- 13 years; 82% males; 59% with CHF; 59% with prior MI) evaluated for cardiac disease, CMR data were analyzed along with automated and manual calculations of BP data. There was no difference among global LV EF values (ANOVA p = 0.90) and strong correlation between automated (r = 0.96, p < 0.0001) and manual (r = 0.95, p < 0.0001) global LV EFs versus CMR, with no significant trends or biases. There was "very good agreement" of automated (kappa = 0.91) and manual (kappa = 0.86) discriminations of cases with LV EF < 50% versus CMR. Detection of LV segments with abnormal regional wall motion was equally accurate (p = 0.68) for automated and manual processings of BP data [ROC areas = 87(+/- 2%) versus 86(+/- 2%)]. The authors conclude that automated and manual computations were equivalent to each other and accurate at identifying both global and regional LV wall motion abnormalities.